MINTEval: Benchmark for Memory in Multi-Target Agent Systems

MINTEval is an analytical benchmark for evaluating memory under multi-target interference in long-horizon agent systems. It was introduced in a research paper and published on Hugging Face by the author 'dinobby' on 2026-05-19. Each example presents a sequence of contexts followed by questions that require reasoning over that history, covering four domains: state tracking, multi-turn dialogue, Wikipedia revisions, and GitHub commits.

Use Cases

• Benchmarking agent memory performance based on sequences of events and edits.
• Evaluating reasoning over long conversation histories based on multi-turn dialogue contexts.
• Testing an AI's ability to track changes and answer questions based on Wikipedia revision histories.
• Assessing memory for code evolution and commit histories in software development contexts.

Strengths

• Covers four distinct domains (state tracking, multi-turn dialogue, Wikipedia revisions, GitHub commits) for varied evaluation.
• Designed specifically for evaluating memory under multi-target interference, a defined research challenge.
• Introduced in a dedicated research paper, suggesting a formal methodological foundation.

Limitations

• Description metadata is limited; actual data quality, size, and structure require manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face dataset uploaded by author 'dinobby'.

Collection Method

Introduced as a benchmark in a research paper; specific data collection method is not detailed in the provided input.

Freshness

Last updated 2026-05-19 15:56:00; freshness should be verified.